It is a bit difficult to know exactly what you are asking. However, point-light recordings are typically perceived as 3D images despite only showing 2D information. Depending on the point-lights shown, a frontal facing point-light walker can be viewed as either approaching or withdrawing and this is determined by the perceivers predisposed bias.

However, you may possibly be asking about using full light video recordings of a person walking and measuring some aspect of their gait. If so, there are a couple of ways it was done before mocap systems were relatively affordable. One way was to attach small lights/reflectors to the joints and record the video in a dark room or turn down the brightness of the video so only the lights are still visible. Another way is to put marks onto the walkers body in a way where they are always visible to the camera as the move in the sagittal or frontal planes. You can then actively measure the angles at each frame of the video. There are phone apps these days that do this type of measurement from videos that were recorded on the phone (e.g. Coach's Eye).

Thanks, Dr. Halovic. Your description involves doing some setup before observation. But in my case, I could only work on pre-recorded videos of gait in static background; CASIA-B dataset to be precise. Is there an alternative method for that?

http://photomodeler.com/index.html

for a few points and a low sample rate, the 3-D reconstruction can be done with comparable simple software. however, this requires a sufficient number of different view angles. in gait analysis you would need a bunch of synchronized cameras the take pictures (high speed around 10 frames per second, or video) from various view angles at the same time. it does even work with consumer electronics.